A new method for enhancing the spatial resolution of lowresolution digital elevation models (DEMs) using a single higherresolution multispectral image is described. The method uses a nonlinear estimation algorithm to determine an empirical relationship between the spectral response of a pixel and the gradient or slope of its underlying elevation surface. This relationship is, in effect, the inverse of the reflectance map used in shape-from-shading to recover surface orientation from image brightness. However, unlike most shape-from-shading techniques, we do not assume constant albedo, nor do we assume a particular functional form for the reflectance map. A depth-from-gradient algorithm is used to generate high-resolution elevations from highresolution gradients using the low-resolution elevations as boundary values. Experimental results demonstrate the ability to generate enhanced resolution elevation maps from global 1-km DEMs and Landsat Thematic Mapper imagery.